Is this image AI generated?
Drop an image to read the labels AI tools and cameras leave inside the file: Content Credentials, IPTC tags, stored prompts and camera data. Everything is read on your device.
JPG, PNG, WebP, AVIF or HEIC. Read on your device, never uploaded.
Choose an imageHow the check works
Drop or paste an image
Drag a file onto the scanner, choose one, or paste it with Ctrl+V. The file is read in your browser and never uploaded.
Every layer is read
The scanner looks for Content Credentials, XMP and IPTC tags, PNG text chunks, EXIF camera data and embedded comments.
Read the verdict
The result says what the file declares about itself, from AI generated to no record at all.
Check the evidence
Each finding shows where it came from and how much weight it carries. Copy the report if you need to share it.
The labels this tool reads
An image file is more than pixels. It can carry several blocks of metadata, each written by different software for a different reason. These are the ones that say something about where an image came from.
| Label | Where it lives | Who writes it | What it tells you |
|---|---|---|---|
| Content Credentials (C2PA) | A signed manifest inside JPEG, PNG, WebP, AVIF and HEIF files | OpenAI, Adobe Firefly, Google Nano Banana Pro, some cameras | Which tool made the file, who signed it, and whether a step was generative AI |
| IPTC digital source type | XMP metadata | Meta AI and software that follows IPTC guidance | trainedAlgorithmicMedia means fully AI generated, compositeWithTrainedAlgorithmicMedia means AI edits inside a real image |
| Generator settings | PNG text chunks and EXIF comments | Stable Diffusion interfaces such as AUTOMATIC1111, ComfyUI, InvokeAI and NovelAI | The prompt, seed, sampler and model used to create the image |
| Camera data | EXIF | Cameras and phones | Camera model, lens and exposure. Easy to copy or type in, so only a hint |
| Editing software | EXIF and XMP | Photoshop, Lightroom, GIMP and other editors | Which apps saved the file along the way |
| Invisible watermarks | Hidden in the pixels | Google SynthID and others | Only the maker's own detector can read them, so this page cannot |
What each result means
Who labels their AI images
OpenAI began adding C2PA metadata to images generated in ChatGPT and its API in February 2024, so anyone can see that an image came from its products. It also warns that this metadata can be removed by platforms, editing tools and file conversions.
Adobeautomatically applies Content Credentials to images made entirely with Firefly, such as Text to Image. Those credentials name Adobe as the issuer and record the AI tool used, and they may be kept in Adobe's public Content Credentials cloud, which is why some files only link to their credentials instead of carrying them.
Google hides an invisible SynthID watermark in images from its AI models. Since November 2025 the Gemini app can check an uploaded image for it, and images made with Nano Banana Pro in the Gemini app, Vertex AI and Google Ads also include C2PA metadata.
Meta marks photorealistic images made with Meta AI in three ways: a visible label, an invisible watermark, and IPTC metadata with the digital source type set to trainedAlgorithmicMedia.
Local Stable Diffusion setups are often the easiest to spot. Popular interfaces such as AUTOMATIC1111 and ComfyUI save the full prompt, seed and settings, or the whole node workflow, into every PNG they write, unless the user turns it off.
Why a clean result proves nothing
Metadata travels with a file only as long as every app along the way keeps it. A screenshot creates a brand new file with none of it. Many social networks and messaging apps strip metadata on upload, and most editors and converters, including the tools on this site, write a fresh file without the old labels. A "no record" result is therefore the most common outcome for images found online, whether they came from a camera or a generator.
When the file has nothing to say, look at the image and its source instead:
- Search for the image with Google Lens or TinEye to find the earliest copy and who posted it first.
- Read any text in the picture. Signs, labels and screens are often garbled in generated images.
- Check hands, teeth, ears, jewelry and glasses, where the two sides of a face often disagree.
- Follow reflections and shadows. They should agree with each other and with the light source.
- Look at the background for melting shapes, repeated faces or objects that merge into each other.
- For images that may come from Google AI, upload them to the Gemini app and ask whether they were made with Google AI.
None of these checks is proof on its own either. Newer generators make fewer of these mistakes every year, so the source of an image usually tells you more than its details.
If you publish AI images
Some platforms require the label to stay. Google Merchant Center, for example, asks that every product image created with generative AI keep the IPTC digital source type set to trainedAlgorithmicMedia, and tells sellers not to remove it. Upload the original file from the generator in that case, and check it here first to confirm the label is present. Resizing or compressing it with a canvas based tool, including the ones on this site, will remove the label.
To see or remove the rest of a photo's metadata, such as GPS location, use the EXIF viewer, and read our guide to photo metadata and privacy.
From our researchHow to Tell If an Image Is AI Generated in 2026We ran AI labelled files through 10 everyday edits to see which ones kept the label.